Finance Newsletter Writer Builds AI Research Assistant, Highlights Data Advantage Gap
A veteran financial newsletter writer has spent the past week using Anthropic's Claude Code to build a custom AI assistant that scans daily news and flags relevant stories—a project that took him 40 years to attempt after his last programming effort in the 1980s.
Marc Rubinstein, who publishes Net Interest to finance professionals, deployed the AI coding tool to index his archive of 250-plus newsletter issues spanning five years, then programmed it to monitor financial news sources and surface on-brand content each morning. The experiment worked: "It's been remarkably well attuned," he wrote in his February 20 newsletter, describing how the model learned his editorial patterns from the back catalog.
The anecdote illustrates a broader tension emerging in financial data markets. While Rubinstein benefited from owning a substantial proprietary dataset—his own writing—exchange operators are questioning whether the "data revolution" they've been chasing for years was ever the goldmine promised. "For years, I was told Euronext is missing the data revolution. Data is the new oil," Euronext CEO Stéphane Boujnah said this month. "All of us are finding out that maybe we missed the data boat—that maybe this data boat was a Titanic boat."
Rubinstein's success with Claude Code required minimal technical expertise. His last programming project was a history quiz game in the 1980s, constrained by scarce computer memory. For decades since, he relied on Excel and standard software tools—Google Docs for drafting, Substack for email management, Stripe for payments. The AI assistant filled a gap his existing stack couldn't address: systematic idea generation, the question readers ask him most frequently.
The tool has limits. Rubinstein noted it cannot yet access information locked in private repositories like personal experiences or network contacts, nor can it replicate his writing voice. "Could I add a module to get Claude to write the whole thing for me? Perhaps," he wrote, before listing the constraints.
Adoption of AI coding assistants remains nascent even among tech-savvy users. According to Anthropic, the median Claude Code session lasts just 45 seconds, and only 0.1% of users engage for longer than 40 minutes. The active user base may number no more than one million. Within regulated financial services, uptake appears slower still. A survey of 150 quantitative analysts and researchers suggests the finance community lags the broader tech sector in deployment.
The contrast between Rubinstein's quick win and Boujnah's skepticism about data monetization points to a key variable: proprietary datasets that AI models can learn from. Rubinstein had five years of structured financial commentary. Exchanges have market data, but the value proposition for selling it to AI companies—versus traditional financial firms—remains unclear.
For finance leaders evaluating AI tools, the lesson may be that success depends less on the sophistication of the model than on the quality and ownership of the data it processes. The software Rubinstein built in a week works because he controls the training corpus. The data boat Boujnah describes may have sunk because exchanges don't own the narrative the same way.
Responses (0 )